Context size & token usage calculation seems to be incorrect

[pkhara31]

With a simple "Hi" message sent as the very first message and without any other active chats, the token usage shows 9.2K tokens used.

Also, when set the context to 4K from the existing 65536, I get an error

This is my entire conversation history:

_hi
Hello! How can I help you today?

why the token usage shows 9.2k tokens for a simple hi message ?
The 9.2k token usage for a simple "hi" message is likely due to:

Context overhead: The system includes metadata, conversation history, and internal processing that counts toward token usage, even for simple greetings.

Model-specific counting: Different tokenization methods count tokens differently. Some models count every character or word, while others use subword tokenization (like BPE - Byte Pair Encoding).

System prompts and instructions: The system prompt and any default instructions you've set up are included in the token count.

API overhead: Some APIs add overhead for authentication, routing, and other infrastructure costs.

Batch processing: If you're using batch inference or have context from previous conversations, those tokens are included.

The exact breakdown would depend on your specific model configuration and API settings. Would you like me to help you optimize token usage or understand your specific tokenization setup?_

For this history, the total token usage is 18.1K

[MiaomiaoMaIntel]

@pkhara31 Those 9.2k tokens aren't your message — they're the briefing the SuperClaw reads before it can answer anything. Every message arrives with its instructions (who it is, safety rules, how to handle your files) plus the manuals for the ~30 tools it can use. That's about 9,500 tokens, and it's the same whether you type "hi" or a full page.

On the number growing: that counter is a running total for the whole conversation, not per message. After your first message it read 9.2k; after your second it read 18.1k — so the second message was about 9k, the same as the first, not extra. Why it counts every time: superclaw token usage count does not count cache hit. It shows up in the count each turn even though the system keeps a cached copy and doesn't reprocess it from scratch — that's why replies get faster after the first.

This is normal, and we're on the lean side. Anthropic's own Claude Code carries about 20k of the same kind of overhead — roughly double.
You weren't charged anything. This is the local model running on your own machine, so there's no per-token billing at all. The count affects speed, not cost.